Automation workflows

Use a workflow for repeatable, read-only checks across one or more IFC files. No console, server, or LLM is required.

manifest -> plan -> reviewed plan -> run -> artifact manifest
              no work             jobs and batches

Version 1 supports schema/IDS validation and selector queries.

Example manifest

version: "1"
name: submission-gate

inputs:
  - id: models
    paths: [models/*.ifc]

failure_policy: continue

steps:
  - id: validate
    input_ids: [models]
    concurrency: 2
    operation:
      kind: validation
      version: "1"
      ids_paths: [requirements/submission.ids]

  - id: external-walls
    input_ids: [models]
    needs: [validate]
    operation:
      kind: query
      version: "1"
      query: IfcWall, Pset_WallCommon.IsExternal=TRUE
      fields: [name, storey, type_name]
      output_format: csv

Paths and globs are relative to the manifest. Absolute paths and .. are rejected. Every pattern must match at least one file. Step IDs, output names, and dependencies must be valid and unique; cycles and unknown keys fail before work starts.

step_concurrency limits running steps. A step's concurrency limits its child jobs. failure_policy accepts continue or fail_fast.

Plan and run

ifc-console workflows schema > workflow-v1.schema.json
ifc-console run workflow.yaml --plan --json
ifc-console run workflow.yaml --output-dir reports

Planning resolves paths, hashes every IFC and IDS source, checks policy, and produces a stable plan_id. It creates no jobs or artifacts.

Running verifies the captured inputs again, stores the workflow record, then schedules ready steps. Exit codes:

code	meaning
0	every step ran and validation passed
5	work ran, but validation findings failed the gate
1	source, worker, batch, workflow, or cancellation failure

Manage runs

ifc-console workflows list
ifc-console workflows show workflow-0123456789abcdef
ifc-console workflows watch workflow-0123456789abcdef
ifc-console workflows cancel workflow-0123456789abcdef
ifc-console workflows resume workflow-0123456789abcdef

Records store steps, dependencies, attempts, batches, failures, and artifacts. Resume re-hashes inputs, verifies completed artifacts, reuses only valid work, and retries unfinished steps. Changed source bytes require a new plan.

The final content-addressed manifest references every step and child report, so artifact cleanup retains the complete result graph.

Python

from ifc_console import Workbench

with Workbench.open(home=".ifc-console-ci") as wb:
    plan = wb.plan_workflow("workflow.yaml")
    run = wb.submit_workflow_plan(plan)

    for update in wb.watch_workflow(run.workflow_id):
        print(update.progress, update.message)

    completed = wb.workflow(run.workflow_id)

submit_workflow() plans and submits in one call. The SDK also provides list, wait, cancel, and resume methods. AsyncWorkbench exposes async equivalents.

examples/workflows/submission-gate.yaml is a copyable template.

Security boundary

Version 1 cannot run generated Python, shell commands, plugins, network calls, or model mutations. YAML uses safe loading; manifests are size limited; paths stay inside allowed roots; and source counts, hashing, worker time, memory, filesystem, network, and subprocess access are bounded.

Manifests do not interpolate environment variables or secrets.